A company manufactures "high thermal performance PVC window frames". Its website repeats it everywhere, and so do its campaigns. Meanwhile, in real conversations, people write: "I need to change my windows, the house is freezing". They search "new windows tax deduction", they ask ChatGPT "how much does it cost to replace the windows in a 90-square-metre flat". Nobody, anywhere in the observable world, has ever spontaneously typed "high thermal performance window frames".
This scene repeats identically in almost every industry we analyse. The company names the product with its internal vocabulary — the technical office, the price list, the trade fair. The recipient names the problem with the vocabulary of their own life. And every time the two vocabularies fail to meet, the company becomes invisible at the exact moment someone is looking for what it sells.
Studying your audience is not an academic indulgence. It is the only insurance against talking to yourself.
The good news: audiences have never been so observable — reviews, forums, communities, threads, videos, questions asked to voice assistants and language models. A vast, public, real-time ethnographic archive. The bad news: most companies still study it with a tool born thirty years ago — personas.
Personas were born in late-'90s software design with a sensible intuition: designing for a concrete archetype beats designing for "the average user". In the passage from design to marketing, though, that intuition degenerated into a ritual: the slide with the stock photo, the invented name, the age, the "pain points" written in a meeting room by people who never observed the actual recipient.
The flaw is not cosmetic — it is structural. The traditional persona is demographic (it describes who you are, not what you are trying to solve), static (produced once, ageing in a PDF), and self-referential (it portrays the customer the company imagines, not the one that exists). Two people with the same age, income and city can hold opposite intentions in front of the same product; the same person changes intention three times in a week. Demographics cannot see this. Behaviour can.
The point is not to throw away the idea of the archetype. It is to change its raw material: replace projection with observation. And there is a discipline that has made a rigorous method of observation for over a century.
Ethnography is the study of communities through direct, prolonged observation: you stay inside the context, you listen to how people talk to each other, you map what they do — not what they claim to do when interviewed. Digital anthropology transfers this method to the spaces where communities now live a substantial part of their lives: groups, forums, reviews, comments, subreddits, vertical communities.
For a business, the field of observation is the community of reference: the set of people who share the problem the product solves, and who talk about it with each other. People renovating a house have their places, their linguistic rituals, their hierarchies of trust ("the heating engineer on the forum said…"). People running a gym, buying ERP software, looking for a lawyer: every community has a lexicon, recurring anxieties, decision criteria that no questionnaire surfaces — because in questionnaires people answer, in communities they talk.
Observing a community of reference ethnographically means three very concrete things. First: mapping the contexts — where the conversations that matter happen, who holds authority there, which sources get quoted. Second: mapping the words — how people name what the company sells, with which synonyms, distortions, brand names used as common nouns. Third: mapping the tensions — the points where desire meets a fear, a doubt, a friction ("I'd like to, but…"). Tensions are the most precious material: that is where purchases are decided, and where almost no corporate communication ever goes to work.
This work on language has an immediate operational consequence, because the way people search has changed three times in ten years — and each change made the observable material richer.
First era: keywords
For twenty years the unit of measure was the keyword: two or three words typed into a box ("pvc windows prices"). Useful, but poor: a keyword says what, not why now, not with what anxiety.
Second era: search intent
Mature SEO stopped counting keywords and started classifying intent: researching, comparing, buying, solving an urgent problem. The same word carries different intents — "boiler" at 3pm in July is an evaluation, at 11pm in January it is an emergency — and communicating well means answering the intent, not the word.
Third era: full sentences, spoken aloud and written to AI
With voice assistants first and language models now, people have stopped compressing themselves into keywords. Nobody writes "pvc windows naples" to ChatGPT: they write "I live in Naples in a 1960s building, I have quotes ranging from 8 to 15 thousand euros to replace 7 windows, how do I know if I'm being ripped off?". A sentence like that contains the context, the budget, the tension, the decision criterion. It is ethnographic gold, handed over spontaneously — and companies still thinking in keywords lack even the conceptual tools to read it.
The keyword was an X-ray. The sentence spoken to an AI is a confession.
If traditional personas are demographic projections, the enriched persona is a profile built from observational data. It does not start from "who they are" but from "what they do, what they say, what they want, what they fear". In our method it is built with five layers of analysis, each fed by real data.
1. Data retrieval
The systematic collection of traces: public conversations, reviews, threads, search queries, questions asked to AI, customer-care transcripts. The corpus must be built with ethnographic judgement — covering the places where the community of reference actually talks, not just the ones easy to scrape.
2. Behaviour analysis
What people do: journeys, entry and abandonment points, seasonality, comparison rituals ("I get three quotes", "I read the negative reviews first"). Observed behaviour systematically corrects what people declare.
3. Intent analysis
Classifying what people are trying to achieve at each moment of the journey: understand, compare, validate a choice already made, find social proof, escape an emergency. Each intent requires different content, a different tone, a different channel.
4. Tension analysis
The conflicts that hold decisions back: desire versus suspicion ("I like it, but it looks too cheap to be true"), urgency versus distrust, aspiration versus social judgement. Tensions only emerge from listening to conversations — and it is exactly what our Emotional Friction Score™ measures in RADAR.
5. Conversation analysis
The linguistic layer: the exact words, the metaphors, the recurring phrases, what gets asked and what gets left unsaid. This produces the real dictionary of the community of reference — the one the communication will have to be written in.
The result is not a slide: it is a living profile, refreshed at every analysis cycle, answering the operational questions — which words to be found by, which objections to dissolve before they are voiced, which tone to use, where to be present.
All this work has one purpose: designing communication tools in people's own language, aligned with their real desires. When the enriched persona is done well, downstream choices stop being opinions. The website uses the words of the community of reference, not those of the price list. Content answers intent, in the form it is expressed — including the questions asked to AI. Campaigns work on real tensions instead of repeating product features. And sometimes the offer itself gets corrected: because listening reveals that the real problem was not quite the one the company thought it was solving.
It is the difference between communicating at people and communicating with people: solutions that solve real problems, told with the words of those who live them. Everything else — brand awareness, positioning, even ROAS — comes after, and comes better.
The method described is the one applied in RADAR, goodea's proprietary market intelligence: analysis of millions of public conversations per quarter, with dedicated indicators for message resonance, emotional friction and promise-reality gap. For an analysis of your community of reference, write to commerciale@goodea.it.